feat: add LiteLLM support

aider/coders/base_coder.py

@@ -73,7 +73,7 @@ def create(
 
  if not skip_model_availabily_check and not main_model.always_available:
  if not check_model_availability(io, client, main_model):
- fallback_model = models.GPT35_0125
+ fallback_model = models.Model.create("gpt-3.5-turbo-0125")
  io.tool_error(
  f"API key does not support {main_model.name}, falling back to"
  f" {fallback_model.name}"
@@ -160,7 +160,7 @@ def __init__(
  if use_git:
  try:
  self.repo = GitRepo(
- self.io, fnames, git_dname, aider_ignore_file, client=self.client
+ self.io, fnames, git_dname, main_model, aider_ignore_file, client=self.client
  )
  self.root = self.repo.root
  except FileNotFoundError:
@@ -224,7 +224,7 @@ def __init__(
 
  self.summarizer = ChatSummary(
  self.client,
- models.Model.weak_model(),
+ self.main_model.get_weak_model(),
  self.main_model.max_chat_history_tokens,
  )
 
@@ -1055,6 +1055,9 @@ def dirty_commit(self):
 
 
 def check_model_availability(io, client, main_model):
+ if not hasattr(client, "models"):
+ return True
+
  try:
  available_models = client.models.list()
  except openai.NotFoundError:

aider/main.py

@@ -13,6 +13,7 @@
 from aider.io import InputOutput
 from aider.repo import GitRepo
 from aider.versioncheck import check_version
+from aider.models.litellm import LITELLM_SPEC
 
 from .dump import dump # noqa: F401
 
@@ -163,7 +164,6 @@ def main(argv=None, input=None, output=None, force_git_root=None):
  core_group.add_argument(
  "--model",
  metavar="MODEL",
- default=default_model,
  help=f"Specify the model to use for the main chat (default: {default_model})",
  )
  core_group.add_argument(
@@ -190,16 +190,16 @@ def main(argv=None, input=None, output=None, force_git_root=None):
  const=default_4_turbo_model,
  help=f"Use {default_4_turbo_model} model for the main chat",
  )
- default_3_model = models.GPT35_0125
+ default_3_model = "gpt-3.5-turbo-0125"
  core_group.add_argument(
  "--35turbo",
  "--35-turbo",
  "--3",
  "-3",
  action="store_const",
  dest="model",
- const=default_3_model.name,
- help=f"Use {default_3_model.name} model for the main chat",
+ const=default_3_model,
+ help=f"Use {default_3_model} model for the main chat",
  )
  core_group.add_argument(
  "--voice-language",
@@ -247,6 +247,11 @@ def main(argv=None, input=None, output=None, force_git_root=None):
  const="https://openrouter.ai/api/v1",
  help="Specify the api base url as https://openrouter.ai/api/v1",
  )
+ model_group.add_argument(
+ "--litellm",
+ action="store_true",
+ help="Use litellm instead of openai",
+ )
  model_group.add_argument(
  "--edit-format",
  metavar="EDIT_FORMAT",
@@ -543,6 +548,8 @@ def main(argv=None, input=None, output=None, force_git_root=None):
  check_gitignore(git_root, io)
 
  def scrub_sensitive_info(text):
+ if not args.openai_api_key:
+ return text
  # Replace sensitive information with placeholder
  return text.replace(args.openai_api_key, "***")
 
@@ -557,18 +564,73 @@ def scrub_sensitive_info(text):
 
  io.tool_output(*map(scrub_sensitive_info, sys.argv), log_only=True)
 
- if not args.openai_api_key:
- if os.name == "nt":
- io.tool_error(
- "No OpenAI API key provided. Use --openai-api-key or setx OPENAI_API_KEY."
- )
- else:
- io.tool_error(
- "No OpenAI API key provided. Use --openai-api-key or export OPENAI_API_KEY."
- )
+ if args.litellm:
+ if not args.model:
+ io.tool_error("You must specify --model or AIDER_MODEL environment variable when using --litellm.")
+ return 1
+
+ elif not (args.openai_api_key or args.openai_api_base) and \
+ args.model is not None and \
+ LITELLM_SPEC is not None:
+ io.tool_output(f"OpenAI key not provided, using LiteLLM instead.")
+ args.litellm = True
+
+ elif not args.openai_api_key:
+ export_kw = "setx" if os.name == "nt" else "export"
+ io.tool_error(
+ f"No OpenAI API key provided. Use --openai-api-key or {export_kw} OPENAI_API_KEY."
+ )
  return 1
 
- if args.openai_api_type == "azure":
+ if not args.model:
+ args.model = default_model
+
+ if args.litellm:
+ if LITELLM_SPEC is None:
+ io.tool_error("LiteLLM is not installed. Install it with `pip install litellm`.")
+ return 1
+
+ io.tool_output("LiteLLM is enabled.")
+
+ packageRootDir = os.path.dirname(LITELLM_SPEC.origin)
+ modelPricesBackupName = "model_prices_and_context_window_backup.json"
+ modelPricesPath = os.path.join(packageRootDir, modelPricesBackupName)
+
+ # LiteLLM keeps a backup of the model prices for when the network is
+ # down, but it's otherwise not used unless LITELLM_LOCAL_MODEL_COST_MAP
+ # is set to True. To keep Aider's startup time fast, prevent the network
+ # request unless the model prices haven‘t been updated in 12 hours.
+ from datetime import datetime, timedelta
+ timeSinceModified = datetime.now() - datetime.fromtimestamp(os.path.getmtime(modelPricesPath))
+
+ if timeSinceModified < timedelta(hours=12):
+ os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+ io.tool_output("LiteLLM model list backup is recent, using it instead of network request.")
+ else:
+ def strfdelta(delta):
+ days = delta.days
+ hours, remainder = divmod(delta.seconds, 3600)
+ minutes, seconds = divmod(remainder, 60)
+ time_parts = [f"{days} day{'' if days == 1 else 's'}" if days else "",
+ f"{hours} hour{'' if hours == 1 else 's'}" if hours else "",
+ f"{minutes} minute{'' if minutes == 1 else 's'}" if minutes else "",
+ f"{seconds} second{'' if seconds == 1 else 's'}"]
+ time_parts = [part for part in time_parts if part]
+ return ", ".join(time_parts)
+
+ formattedTime = strfdelta(timeSinceModified)
+ io.tool_output(f"LiteLLM model list backup last updated {formattedTime} ago, updating now.")
+
+ # Since LiteLLM never updates its backup after hitting the
+ # network, we'll have to do it ourselves.
+ from litellm import model_cost
+ with open(modelPricesPath, "w") as f:
+ import json
+ json.dump(model_cost, f)
+
+ from litellm import LiteLLM
+ client = LiteLLM()
+ elif args.openai_api_type == "azure":
  client = openai.AzureOpenAI(
  api_key=args.openai_api_key,
  azure_endpoint=args.openai_api_base,

aider/models/__init__.py

@@ -2,16 +2,10 @@
 from .openai import OpenAIModel
 from .openrouter import OpenRouterModel
 
-GPT4 = Model.create("gpt-4")
-GPT35 = Model.create("gpt-3.5-turbo")
-GPT35_0125 = Model.create("gpt-3.5-turbo-0125")
-
 DEFAULT_MODEL_NAME = "gpt-4-1106-preview"
 
 __all__ = [
+ Model,
  OpenAIModel,
  OpenRouterModel,
- GPT4,
- GPT35,
- GPT35_0125,
 ]

aider/models/litellm.py

@@ -0,0 +1,67 @@
+import importlib.util
+import logging
+import tiktoken
+
+from .model import Model
+
+logging.basicConfig(level=logging.INFO, format="%(message)s")
+logger = logging.getLogger("aider-litellm")
+
+LITELLM_SPEC = importlib.util.find_spec("litellm")
+
+model_aliases = {
+ # claude-3
+ "opus": "claude-3-opus-20240229",
+ "sonnet": "claude-3-sonnet-20240229",
+ "haiku": "claude-3-haiku-20240307",
+ # gemini-1.5-pro
+ "gemini": "gemini/gemini-1.5-pro-latest",
+ # gpt-3.5
+ "gpt-3.5": "gpt-3.5-turbo-0613",
+ "gpt-3.5-turbo": "gpt-3.5-turbo-0613",
+ "gpt-3.5-turbo-16k": "gpt-3.5-turbo-16k-0613",
+ # gpt-4
+ "gpt-4": "gpt-4-0613",
+ "gpt-4-32k": "gpt-4-32k-0613",
+}
+
+models_info = None
+
+class LiteLLMModel(Model):
+ def __init__(self, name):
+ model_id = name
+ if name in model_aliases:
+ model_id = model_aliases[name]
+
+ from litellm import model_cost
+
+ model_data = model_cost.get(model_id)
+ if not model_data:
+ # For gemini 1.5 pro to work, LiteLLM appears to need the "-latest"
+ # part included in the model name, but it's not included in the list
+ # of supported models that way, so finesse it here
+ if model_id == "gemini/gemini-1.5-pro-latest":
+ model_data = model_cost.get("gemini/gemini-1.5-pro")
+ if not model_data:
+ raise ValueError(f"Unsupported model: {model_id}")
+ else:
+ raise ValueError(f"Unsupported model: {model_id}")
+
+ self.tokenizer = tiktoken.get_encoding("cl100k_base")
+
+ self.name = model_id
+ self.max_context_tokens = model_data.get("max_input_tokens")
+ self.prompt_price = model_data.get("input_cost_per_token") * 100
+ self.completion_price = model_data.get("output_cost_per_token") * 100
+
+ is_high_end = model_id.startswith("gpt-4") or model_id.startswith("claude-3-opus")
+
+ self.edit_format = "udiff" if is_high_end else "whole"
+ self.use_repo_map = is_high_end
+ self.send_undo_reply = is_high_end
+
+ # set the history token limit
+ if self.max_context_tokens < 32 * 1024:
+ self.max_chat_history_tokens = 1024
+ else:
+ self.max_chat_history_tokens = 2 * 1024

aider/models/model.py

@@ -22,7 +22,10 @@ class Model:
  def create(cls, name, client=None):
  from .openai import OpenAIModel
  from .openrouter import OpenRouterModel
+ from .litellm import LiteLLMModel
 
+ if client and not hasattr(client, "base_url"):
+ return LiteLLMModel(name)
  if client and client.base_url.host == "openrouter.ai":
  return OpenRouterModel(client, name)
  return OpenAIModel(name)
@@ -38,9 +41,10 @@ def strong_model():
  def weak_model():
  return Model.create("gpt-3.5-turbo-0125")
 
- @staticmethod
- def commit_message_models():
- return [Model.weak_model()]
+ def get_weak_model(self):
+ if self.use_repo_map:
+ return Model.weak_model()
+ return self
 
  def token_count(self, messages):
  if not self.tokenizer:

aider/repo.py

@@ -16,7 +16,8 @@ class GitRepo:
  aider_ignore_spec = None
  aider_ignore_ts = 0
 
- def __init__(self, io, fnames, git_dname, aider_ignore_file=None, client=None):
+ def __init__(self, io, fnames, git_dname, main_model=None, aider_ignore_file=None, client=None):
+ self.main_model = main_model
  self.client = client
  self.io = io
 
@@ -120,10 +121,8 @@ def get_commit_message(self, diffs, context):
  dict(role="user", content=content),
  ]
 
- for model in models.Model.commit_message_models():
- commit_message = simple_send_with_retries(self.client, model.name, messages)
- if commit_message:
- break
+ commit_model = self.main_model.get_weak_model()
+ commit_message = simple_send_with_retries(self.client, commit_model.name, messages)
 
  if not commit_message:
  self.io.tool_error("Failed to generate commit message!")